GMRT Archive Processing Project

TL;DR

The GMRT Archive Processing Project aims to develop a pipeline to process and deliver large volumes of GMRT radio telescope data, enhancing scientific research and enabling systematic performance analysis.

Contribution

This work introduces a large-scale, uniform data processing pipeline for GMRT observations across multiple frequencies, providing accessible, science-ready data products.

Findings

Processed thousands of images across multiple bands

Enabled studies of galaxy clusters, AGN, and star-forming regions

Systematic analysis of telescope performance characteristics

Abstract

The GMRT Online Archive now houses over 120 terabytes of interferometric observations obtained with the GMRT since the observatory began operating as a facility in 2002. The utility of this vast data archive, likely the largest of any Indian telescope, can be significantly enhanced if first look (and where possible, science ready) processed images can be made available to the user community. We have initiated a project to pipeline process GMRT images in the 150, 240, 325 and 610 MHz bands. The thousands of processed continuum images that we will produce will prove useful in studies of distant galaxy clusters, radio AGN, as well as nearby galaxies and star forming regions. Besides the scientific returns, a uniform data processing pipeline run on a large volume of data can be used in other interesting ways. For example, we will be able to measure various performance characteristics of the GMRT telescope and their dependence on waveband, time of day, RFI environment, backend, galactic latitude etc. in a systematic way. A variety of data products such as calibrated UVFITS data, sky images and AIPS processing logs will be delivered to users via a web-based interface. Data products will be compatible with standard Virtual Observatory protocols.

Full text

1 Introduction

The Giant Meterwave Radio Telescope (GMRT, Swarup et al. 1991) is a low frequency radio interferometer operating at a site 80 km north of Pune, India. Since 2002, it has been operated as an international open access facility by India’s National Centre for Radio Astrophysics. All interferometric observations carried out with the GMRT have been carefully archived over the years using several different tape and disk based storage technologies (see Fig.1). Raw interferometric visibilities were made available to the international user community, on request, via DVDs until 2009. Thereafter, all raw data were made accessible for search and download via a password authenticated, web-based interface, the NCRA Archive and Proposal handling System (NAPS)111http://naps.ncra.tifr.res.in/goa

The NAPS system now hosts over 120 TB of data and delivers them to the GMRT user community located in about 40 countries worldwide. In 2018, we saw an average of about 55 data requests per month with an average size of 50 GB per request. Despite this high level of usage of the archive, the scientific utility of these data is greatly limited by the large effort required to transform these raw visibilities into science ready images. To address this situation, we have initiated an effort to generate pipeline processed continuum images for GMRT data. We are looking to provide users with “first look” (worst case) and “science ready” (best case) images for as many GMRT observations as possible. A “certifiably bad” tag on data is also useful, because it helps convince the time allocation committee that fresh observations are warranted.

2 Imaging the GMRT Archive

Presently, there is no standard or official pipeline for processing GMRT data. However, there are several scripts and pipelines developed by different users, some of which are publicly available. One of the most sophisticated, publicly available, pipelines for processing data from the GMRT is the Source Peeling and Modelling (SPAM) pipeline developed by H. Intema (Intema et al. 2017). It was used to successfully pipeline process about 2000 hours of GMRT data from the TIFR GMRT Sky Survey (TGSS ADR1222http://tgssadr.strw.leidenuniv.nl, Intema et al. 2017). SPAM is a Python module that provides an interface to AIPS via ParselTongue (Kettenis et al. 2006) and ObitTalk (Cotton 2008). ParselTongue provides access to AIPS tasks, data files (images & visibilities) and tables. SPAM extensively uses several other popular Python modules like numpy and scipy. Data reductions are carried out by well-tested Python scripts that executes AIPS tasks directly or via high-level functions that make multiple AIPS or ParselTongue calls. SPAM now also includes a fully automated pipeline for reducing legacy GMRT observations at 150, 235, 325 and 610 MHz.

3 Building our compute infrastructure

We used a simple Beowolf cluster architecture for our processing. A master node acts as a fileserver for the compute stack - AIPS, SPAM, Obit and Parseltongue which is NFS exported to a set of compute nodes (simple headless desktop computers) which have a standard Ubuntu 16.04 server installation plus some additional software libraries installed via Tentakel from the master node. Ganglia was chosen as the tool for monitoring cluster status. After some experimentation, we found that using a Docker container to install our software stack on each compute node was more efficient. We began processing with a four node cluster which was gradually expanded to include about 30 headless desktops. Even with this modest hardware, it is possible to process about 5 months of GMRT data in about a month. The raw data archive is hosted on a Dell EMC Isilon system from where it is NFS exported to our data processing cluster. After processing, the outputs are copied back onto the Isilon system for long term storage and disaster recovery compliant backup.

4 Data processing and delivery

The SPAM pipeline is designed to be used interactively by a single user and each processing instance runs as a single thread on the CPU. We wrote a set of Python and bash scripts to make SPAM operate in non-interactive fashion and to run multiple processing threads simultaneously on each multicore computer in our cluster.

We realised quickly that keeping track of the processing was very cumbersome since data processing rarely progressed linearly. Failures could happen due to poor data quality or due to some limitation in SPAM. It was important to bookkeep all of these so that we could get an accurate picture of the current status of the processing for each observation and to gather statistics on failure situations. We have developed a comprehensive database schema to keep track of the processing. Scattered throughout the SPAM processing are read and write calls to the database recording successes and failure and metadata on them. This database is also critical in determining which outputs are ready to be delivered to users after some automated and manual quality control. The database will also prove useful in analysis of the long term trends at the observatory in terms of the evolving telescope charateristics with manmade radio frequency interference and the ionospheric environment.

For datasets where the processing is successful (see Fig.2) a variety of data products such as calibrated UVFITS data, sky images, AIPS processing logs are generated and are now being integrated into the NAPS system for delivery. These will become visible to NAPS users as additional value added data products which will be compatible with standard Virtual Observatory protocols.

5 Future plans

The current SPAM pipeline only works on legacy GMRT data. For data which is now streaming from the upgraded GMRT (uGMRT, Gupta et al. 2017), with seamless frequency coverage and large bandwidth, a different pipeline would be needed. It would also be tremendously useful if the data processing can be done in near real time, so that authorised users can view the processed images from their own observations, within a few days. We are currently exploring the software and hardware enhancements that are necessary to enable these exciting possibilities over the next few years.